m Ji

Segmentation, Inference and Classification of Partially Overlapping Nanoparticles

This paper presents a method that enables automated morphology analysis of partially overlapping nanoparticles in electron micrographs. In the undertaking of morphology analysis, three tasks appear necessary: separate individual particles from an agglomerate of overlapping nano-objects; infer the particles missing contours; and ultimately, classify the particles by shape based on their complete contours. Our specific method adopts a two-stage approach: the first stage executes the task of particle separation, and the second stage conducts simultaneously the tasks of contour inference and shape classification. For the first stage, a modified ultimate erosion process is developed for decomposing a mixture of particles into markers, and then, an edge-to-marker association method is proposed to identify the set of evidences that eventually delineate individual objects. We also provided theoretical justification regarding the separation capability of the first stage. In the second stage, the set of evidences become inputs to a Gaussian mixture model on B-splines, the solution of which leads to the joint learning of the missing contour and the particle shape. Using twelve real electron micrographs of overlapping nanoparticles, we compare the proposed method with seven state-of-the-art methods. The results show the superiority of the proposed method in terms of particle recognition rate.

An SVDD-Based Algorithm for Target Detection in Hyperspectral Imagery

Spectral variability remains a challenging problem for target detection and classification in hyperspectral (HS) imagery. In this letter, we have applied the nonlinear support vector data description (SVDD) to perform full-pixel target detection. Using a pure target signature and a first-order Markov model, we have developed a novel pattern recognition algorithm to train an SVDD to characterize the target class. We have inserted target signatures into an urban HS scene with varying levels of spectral variability to explore the performance of the proposed SVDD target detector in different scenarios. The proposed approach makes no assumptions regarding the underlying distribution of the scene data as do traditional stochastic detectors such as the matched filter (MF). Detection results in the form of confusion matrices, and receiver-operating-characteristic curves demonstrate that the proposed SVDD-based algorithm is highly accurate and yields higher true positive rates and lower false positive rates than the MF.

Dynamic MRI with compressed sensing imaging using temporal correlations

Compressed sensing (CS) is a recently emerged technique for reconstructing signals from data sampled under the Nyquist rate. It takes advantage of the signal sparsity in a transformed domain to reconstruct high-resolution signals from reduced data. This paper presents a CS imaging method for dynamic magnetic resonance imaging. Specifically, a difference operator is applied to the temporal data frames to enhance the spatial signal sparsity for CS reconstruction. The new algorithm method was assessed using simulated and in-vivo dynamic imaging data. The result shows that the new method can obtain higher resolution than zero-padded Fourier reconstruction and the Keyhole method, and it results in reduced artifacts and noise than conventional CS reconstruction where no temporal information is used. It also shows that the new CS dynamic imaging method does not suffer substantial signal-to-noise loss.

Compressed sensing MRI with multichannel data using multicore processors

Compressed sensing (CS) is a promising method to speed up MRI. Because most clinical MRI scanners are equipped with multichannel receive systems, integrating CS with multichannel systems may not only shorten the scan time but also provide improved image quality. However, significant computation time is required to perform CS reconstruction, whose complexity is scaled by the number of channels. In this article, we propose a reconstruction procedure that uses ubiquitously available multicore central processing unit to accelerate CS reconstruction from multiple channel data. The experimental results show that the reconstruction efficiency benefits significantly from parallelizing the CS reconstructions and pipelining multichannel data into multicore processors. In our experiments, an additional speedup factor of 1.6–2.0 was achieved using the proposed method on a quad‐core central processing unit. The proposed method provides a straightforward way to accelerate CS reconstruction with multichannel data for parallel computation. Magn Reson Med, 2010.

Compressed sensing parallel Magnetic Resonance Imaging

Both parallel Magnetic Resonance Imaging (pMRI) and Compressed Sensing (CS) can significantly reduce imaging time in MRI, the former by utilizing multiple channel receivers and the latter by utilizing the sparsity of MR images in a transformed domain. In this work, pMRI and CS are integrated to take advantages of the sensitivity information from multiple coils and sparsity characteristics of MR images. Specifically, CS is used as a regularization method for the inverse problem raised by pMRI based on the L1 norm and a Total Variation (TV) term. We test the new method with a set of 8-channel, in-vivo brain MRI data at reduction factors from 2 to 8. Reconstruction results show that the proposed method outperforms several other regularized parallel MRI reconstruction such as the truncated Singular Value Decomposition (SVD) and Tikhonov regularization methods, in terms of residual artifacts and SNR, especially at reduction factors larger than 4.

Spectral phase-corrected GRAPPA reconstruction of three-dimensional echo-planar spectroscopic imaging (3D-EPSI).

MR spectroscopic (MRS) images from a large volume of brain can be obtained using a 3D echo‐planar spectroscopic imaging (3D‐EPSI) sequence. However, routine applications of 3D‐EPSI are still limited by a long scan time. In this communication, a new approach termed “spectral phase‐corrected generalized autocalibrating partially parallel acquisitions” (SPC‐GRAPPA) is introduced for the reconstruction of 3D‐EPSI data to accelerate data acquisition while preserving the accuracy of quantitation of brain metabolites. In SPC‐GRAPPA, voxel‐by‐voxel spectral phase alignment between metabolite 3D‐EPSI from individual coil elements is performed in the frequency domain, utilizing the whole spectrum from interleaved water reference 3D‐EPSI for robust estimation of the zero‐order phase correction. The performance of SPC‐GRAPPA was compared with that of fully encoded 3D‐EPSI and conventional GRAPPA. Analysis of whole‐brain 3D‐EPSI data reconstructed by SPC‐GRAPPA demonstrates that SPC‐GRAPPA with an acceleration factor of 1.5 yields results very similar to those obtained by fully encoded 3D‐EPSI, and is more accurate than conventional GRAPPA. Magn Reson Med 57:815–820, 2007.

Reducing SAR in parallel excitation using variable-density spirals : a simulation-based study

Parallel excitation using multiple transmit channels has emerged as an effective method to shorten multidimensional spatially selective radiofrequency (RF) pulses, which have a number of important applications, including B1 field inhomogeneity correction in high-field MRI. The specific absorption rate (SAR) is a primary concern in high-field MRI, where wavelength effects can lead to local peaks in SAR. In parallel excitation, the subjects are exposed to RF pulses from multiple coils, which makes the SAR problem more complex to analyze, yet potentially enables greater freedom in designing RF pulses with lower SAR. Parallel-excitation techniques typically employ either Cartesian or constant-density (CD) spiral trajectories. In this article, variable-density (VD) spiral trajectories are explored as a means for SAR reduction in parallel-excitation pulse design. Numerical simulations were conducted to study the effects of CD and VD spirals on parallel excitation. Specifically, the electromagnetic fields of a four-channel transmit head coil with a three-dimensional head model at 4.7 T were simulated using a finite-difference time domain method. The parallel RF pulses were designed and the resulting excitation patterns were generated using a Bloch simulator. The SAR distributions due to CD and VD spirals were evaluated quantitatively. The simulation results show that, for the same pulse duration, parallel excitation with VD spirals can achieve a lower SAR compared to CD spirals for parallel excitation. VD spirals also resulted in reduced artifact power in the excitation patterns. This gain came with slight, but noticeable, degrading of the spatial resolution of the resulting excitation patterns.

Susceptibility-based positive contrast MRI of brachytherapy seeds.

To provide visualization of the brachytherapy seeds and differentiation with natural structures in MRI by taking advantage of their high magnetic susceptibility to generate positive‐contrast images.

Minimal-SAR RF pulse optimization for parallel transmission in MRI

Parallel transmission is an emerging technique to achieve multi-dimensional spatially selective or modulated excitation in Magnetic Resonance Imaging (MRI). Minimizing Specific Absorption Ratio (SAR) is a critical issue in this technique for radio frequency power absorption safety. In this paper, we presented an automatic design method to minimize SAR in an optimization framework. The RF pulses and corresponding k-space trajectory are iteratively adjusted. The method is verified using computer simulations of a 4-channel parallel transmission system. The results showed significantly reduction in SAR can be achieved while the quality of the excited pattern is well preserved without enlonging the pulse duration.

Quantitative magnetic resonance imaging in a naturally occurring canine model of spinal cord injury.

Study design:Retrospective cohort study.Objectives:To analyze magnetic resonance imaging (MRI) evaluator agreement in dogs with spinal cord injury (SCI) caused by intervertebral disk herniation (IVDH) using semiautomated and manual lesion segmentation and to analyze the associations between MRI and functional outcome.Setting:United States of America.Methods:T2-weighted MRIs from dogs with SCI resulting from thoracolumbar IVDH were identified from a database. Evaluators categorized MRIs on the basis of the presence or absence of a T2-hyperintense spinal cord lesion in axial and sagittal images. A semiautomated segmentation algorithm was developed and used to estimate the lesion volume. Agreement between evaluators and between semiautomated and manual segmentation was analyzed. The relationships of qualitative and quantitative MRIs with behavioral functional outcome were analyzed.Results:Axial images more commonly depicted lesions compared with sagittal images. Lesions in axial images had more consistent associations with functional outcome compared with sagittal images. There was imperfect qualitative agreement, and lesion volume estimation was imprecise. However, there was improved precision using semiautomated segmentation compared with manual segmentation.Conclusion:Lesion volume estimation in dogs with naturally occurring SCI caused by IVDH is challenging, and axial images have important advantages compared with sagittal images. The semiautomated segmentation algorithm described herein shows promise but may require further refinement.